Multi-echelon inventory models for repairable items

We develop multi-echelon inventory models for repairable items and apply them to the management of spare parts inventories. Traditional models focus on military applications by assuming ample repair facilities and large parts populations. As these assumptions are often inappropriate in resource-constrained industrial settings, our aim is to develop improved but simple to implement models relaxing these assumptions, and to propose a comprehensive framework for the selection of appropriate models.; We first place inventory models for repairable items in context by discussing ideas for a practical taxonomy and by discussing the relevance of repairable items. We also provide a comprehensive review of existing models, including the Multi-Echelon Technique for Recoverable Item Control (METRIC) and its variations.; We then propose four new models that relax assumptions of METRIC: (a) Limited repair facilities: We apply an exact Markovian queueing model in conjunction with a double negative binomial approximation. In testing against traditional methods the proposed model outperforms traditional methods, performing especially well under heavy repair utilization, conditions under which the assumption of ample repair capacity introduces serious errors. (b) Limited repair facilities with general repair distributions: We relax the service time distribution assumption of the previous model by using a two-moment queueing model approximation. Experiments with Erlang repair distributions indicate that the approximation is easy to implement and provides good results. (c) Limited repair facilities with multiple classes of customers: The previous model is further relaxed by allowing for different classes of parts in the system, again using a two-moment queueing approximation. Experiments with mixes of Erlang distributions indicate that the approximations are simple to implement and provide good results. (d) Sparse parts population: For settings where the parts population is relatively small we adopt a closed queueing network method, Mean Value Analysis (MVA). We obtain a correction factor which reflects the error in assuming an infinite parts population and can be used to obtain improved performance evaluations.; Based on these results, and a limited industrial survey, an application framework for choosing the appropriate model is proposed. A case study shows the practical use of the framework and the models.